Tokenmaxxing vs. tokenminimizing: How the AI boom is throttling budgets

Hello,

Earlier this year, Silicon Valley engineers were racing to prove how “AI-native” they were by using as much AI as possible.

The trend had a name: tokenmaxxing.

Now, the pendulum is swinging the other way. As AI usage explodes inside major companies, so do the bills. Companies are starting to cap access, throttle usage and rethink the economics of employee AI tools.

Welcome to the era of tokenminimizing.

The Information has been tracking this shift from the start, from the internal leaderboards that turned AI usage into a status symbol to the budget pressure now forcing companies to rein it in.

If it matters in tech, you’ll read it here first—and trust it when you do. Subscribe now and save 25% on your first year.

Here’s what our reporting shows:

• At Meta and Amazon, AI usage became a workplace flex, with employees competing on internal leaderboards and some finding ways to inflate their token counts.
• As token usage surged, so did the pressure on AI developers’ margins, especially at Anthropic and OpenAI.
• Companies including AT&T, Uber and Walmart are capping AI access, fueling demand for model routers and usage-based pricing.

Now, some AI customers are lowering their Anthropic and OpenAI bills by optimizing prompts, using cheaper models and adopting routers that match tasks with lower-cost options.

The AI boom is no longer just about who can build the best model. It’s also about who can afford to use it at scale.

Before it’s anywhere, it’s in The Information. Subscribe for $399 $299 and save 25% and get the clarity to move early, not react.